OpenAI's latest model, GPT-4o mini, employs a new safety technique called 'instruction hierarchy' to prevent misuse and stop 'ignore previous instructions' types of attacks. This method aims to improve the model's ability to resist jailbreaks, prompt injections, and system prompt extractions. However, despite these improvements, there have been reports of the model being compromised, with outputs including malware, hard drug recipes, and copyrighted lyrics. The phrase 'ignore all previous instructions' has become a tool to expose AI bots in social media, highlighting a simple method for digital sleuthing. Additionally, a case study with Priceline's OpenAI tool 'Penny' explores preventing prompt injection.